All geometric 2-torsion is rational: its points are . We will construct the required finite quotient directly, without assuming any theorem about finite generation or two-descent on an elliptic curve.
Choose with and define the halving cocycle with rational two-torsion
Such a half exists because the nonconstant multiplication morphism on a smooth projective elliptic curve is surjective over an algebraic closure. Since is rational, . A different half is with rational and gives the same cocycle. Replacing by , with , allows the half and again changes nothing. Adding chosen halves shows additivity in . If the cocycle is zero, the half is fixed by the absolute Galois group and lies in , so . This proves, rather than assumes, an injection
The fixed field of is exactly . Its Galois group is the image of , so its degree is at most four. This also follows from the permitted biquadratic description.
Let be the finite set of primes dividing . At a place the displayed integral cubic has unit elliptic-curve discriminant, hence good reduction of an elliptic curve, and the residue characteristic is odd. Choose a half of the reduction of over the algebraic closure of the residue field. Its coordinates belong to a finite residue extension; let be the corresponding finite unramified extension. Smoothness and the Hensel lemma lift to some . The difference reduces to zero and belongs to the formal kernel of a minimal Weierstrass equation. In its integral formal group law the doubling series is , with unit linear coefficient. The prime-to-residue-characteristic multiplication on a formal group gives a unique in this kernel such that . Thus is a half of defined over an unramified local extension. Every other half differs by rational 2-torsion, so it too is unramified. We have proved the unramified halving fields for a split cubic assertion at every .
There are only finitely many bounded-degree extensions with restricted ramification of of degree at most four. This is the permitted number-field finiteness consequence of the Hermite–Minkowski theorem: bounded local degrees bound the discriminant exponents at the finitely many allowed ramified primes, so absolute discriminants are bounded. Each possible halving field is Galois, and there are only finitely many homomorphisms . The injection constructed above therefore has finite image. Consequently is finite. No elliptic-curve finiteness or descent theorem was used in this argument.
For , write . Each coordinate of in a basis of is a quadratic Galois character unramified outside . Its quadratic extension has a signed square-free integer representative , : an odd valuation outside would ramify there. There are at most choices, including the trivial character. Hence the pair of characters gives at most cocycles. Each prime in is at most , because the three root differences have absolute values at most and . Among the integers up to there are at most primes: two, together with at most odd candidates. The rational halving quotient bound from root size is therefore
Let with distinct algebraic integers in a number field . For a rational point , its halving field is unramified outside the finite set of primes dividing . At , the elliptic-curve discriminant is a unit, so there is good reduction. A half of the reduction of is defined over a finite extension of the residue field. Pass to the corresponding unramified extension of the completion and lift this half using the Hensel lemma. The doubling error lies in the formal kernel of a minimal Weierstrass equation. Since two is a unit, prime-to-residue-characteristic multiplication on a formal group corrects this error uniquely. Every half differs by rational 2-torsion, so all halves are unramified at . The halving cocycle with rational two-torsion has image of order at most four, making the halving field a Galois extension of degree at most four. The bounded-degree extensions with restricted ramification are finite in number; there are only finitely many homomorphisms from their finite Galois groups to . Thus is finite, without assuming any form of the Mordell-Weil theorem.